I think he's overfitting. Also the way he controls for Lyft seems poor, I think, because if both services led to a reduction them controlling for Lyft would red

I think he's overfitting. Also the way he controls for Lyft seems poor, I think, because if both services led to a reduction them controlling for Lyft would reduce Uber's contribution it seems (that is while Uber's effect might be small as well as Lyfts, together they would be larger). He really should be asking "Do ride sharing services reduce drunk driving fatalities?" Not Uber specific. Also his variable for Uber is only a one or zero. It should instead be the number of drivers in the county. They quotes a study done on only California that shows Uber reduces fatalities, but doesn't ask himself if maybe county by county is important. Instead he groups everything together, even some countries that have very few. Here's the sentence that should indicate something is wrong with his analysis, he finds Uber increases fatalities (a noticably but not significant): > This model shows that on average, the presence of Uber was associated with a 2.0% (95% confidence interval: 0.98, 1.06) increase in traffic fatalities among all drivers; however, this association was not statistically significant at conventional levels. I think he definitely needs to break out the results data into county by county results to see if maybe there is a different result for maybe more populous counties.